Credit: AntonKhrupinArt/Adobe Stock
New study focuses on designing efficient, cost-effective, sustainable engineered systems
Sept 25, 2025
Editor's note: This article originally appeared on Penn State Institute for Computational and Data and Sciences website.
UNIVERSITY PARK, Pa. — Designing complex engineered systems — such as planes, cars and manufacturing systems — with multiple attributes comes with a lot of intricate challenges and a significant amount of computational power, according to Ashwin Renganathan, assistant professor of aerospace engineering, Penn State Institute for Computational and Data Sciences (ICDS) co-hire and director of the Computational complex engineered Systems Design Lab (CSDL) at Penn State.
Renganathan and CSDL graduate student, Kade Carlson, address the challenges of multiobjective optimization, or designing a system with multiple, conflicting attributes that need to be simultaneously optimized under specific constraints. For example, when designing a car optimized for fuel efficiency, the system may need lightweight materials, but these materials may not be safe enough for passengers. Therefore, the design should simultaneously prioritize fuel efficiency and safety.
The study, which was recently published and presented at the 28th International Conference on Artificial Intelligence and Statistics (AISTATS 2025), uses machine learning and Bayesian statistics — a process that takes prior assumptions and updates them as new data becomes available, aiding in better results — to develop a novel optimization algorithm that enables efficient multi-attribute decision-making. The algorithm, called “batch Pareto optimal Thompson sampling (qPOTS)” builds on a classical method of Thompson sampling, which aims to solve multi-armed bandit problems which have multiple options or arms with uncertain results of rewards, similarly to slot machines. qPOTS combines these methods to aid in decision-making processes when designing systems with more than one attribute.
Renganathan also used Penn State’s Roar supercomputer computer processing units (CPUs) and graphics processing units (GPUs) to parallelize — to have more than one computation or experiment working simultaneously — and accelerate experiments, respectively.
For example, qPOTS has been successfully applied to aircraft design.
“Multi-objective optimization is a hard problem to solve,” Renganathan said. “We want to solve problems which are typically solved using derivatives — the rate of change of quality of interest with respect to control parameters of a system — without having access to them. In this case, we wanted to design an airplane for fuel burn efficiency at multiple operating conditions such as flight speeds and altitudes, with respect to the wing shape. The cost of acquiring the sensitivity of the aircraft fuel burn efficiency to the wing shape changes — the derivative — is prohibitive. By using our algorithm, we can arrive at the optimal aircraft design faster and with provable guarantees, without any access to derivatives.”
The team has also open-sourced their qPOTS software, which has been downloaded more than 6,000 times across the world, Renganathan said. He added that his group is actively working on extensions to the algorithm, including high-dimensional scaling which can address systems with several hundreds to thousands of decision variables.
The problem was motivated by many of NASA’s ongoing plans including supersonic commercial transportation and improving efficiencies for long-haul subsonic aircraft, Renganathan said.
“We are designing systems that people can rely on,” he said. “The designs should meet ever-increasing performance goals, while also ensuring safety and sustainability. There’s a lot of uncertainty involved, and computing power is limited. Our algorithms are developed and designed to use minimal computational power. If successful, our tools can bring a shift in the way complex engineered systems are designed. We can improve system performance while maintaining safety and sustainability while having decreased costs and faster turnaround times.”
According to Renganathan, the success of their tools could potentially benefit the broader society by providing access to advanced technologies at an affordable cost.
Share this story:
